Get reshape error for Tensorflow BERT model

[jayjaywg]

Hi,

I converted tensorflow BERT model to onnx by https://github.com/onnx/tensorflow-onnx, but when i do inference, i met error as

GENERAL ERROR : c:\agent_work\3\s\onnxruntime\core\providers\cpu\tensor\reshape_helper.h:43 onnxruntime::ReshapeHelper::ReshapeHelper gsl::narrow_cast<int64_t>(input_shape.Size()) == size was false. The input tensor cannot be reshaped to the requested shape. Input shape:{70,767}

the model have three inputs with shape [100,70], i don't know where did 767 come from? can anyone help?

[yufenglee]

@jayjaywg the input shape:{70,767} is for operator reshape, not the model input. I guess there may be something wrong with the converter. Could you please share the model to @pengwa and @guschmue ? Or log an issue in https://github.com/onnx/tensorflow-onnx instead?

[guschmue]

To convert a tensorflow bert model, follow the instructions here:
https://gist.github.com/guschmue/9f2ba608414bef13fa67cdf278ca1e99

[WilliamTambellini]

Hi @guschmue
Could you please precise what you mean by :
"We want to use placeholders but the TpuEstimator used in run_squad.py doesn't support this,
we require a small change in run_squad.py / model_fn_builder to pass in a argument to a flag
to use TpuEstimator or standard Estimator."
in export_to_saved_model.py ?
Kind

[guschmue]

Sorry, did not get a notify for the question.
This comment in export_to_saved_model.py - we want the input to be a Placeholder when we export the model as saved_model but with the TpuEstimator that is used in bert that doesn't work so in export_to_saved_model.py (which is mostly a copy of run_squad.py) I replaced the TpuEstimator with a normal Estimator. This is the code in if USE_TPU_ESTIMATOR

[WilliamTambellini]

Ok thanks. So perhaps the error I encounter is unrelated to that comment :

INFO:tensorflow:Done calling model_fn.
Traceback (most recent call last):
File "../export_to_saved_model.py", line 113, in
tf.app.run()
File "/usr/local/lib/python3.5/dist-packages/tensorflow/python/platform/app.py", line 125, in run
_sys.exit(main(argv))
File "../export_to_saved_model.py", line 108, in main
serving_input_receiver_fn=serving_input_receiver_fn)
File "/usr/local/lib/python3.5/dist-packages/tensorflow/python/estimator/estimator.py", line 663, in export_savedmodel
mode=model_fn_lib.ModeKeys.PREDICT)
File "/usr/local/lib/python3.5/dist-packages/tensorflow/python/estimator/estimator.py", line 789, in _export_saved_model_for_mode
strip_default_attrs=strip_default_attrs)
File "/usr/local/lib/python3.5/dist-packages/tensorflow/python/estimator/estimator.py", line 907, in _export_all_saved_models
mode=model_fn_lib.ModeKeys.PREDICT)
File "/usr/local/lib/python3.5/dist-packages/tensorflow/python/estimator/estimator.py", line 984, in _add_meta_graph_for_mode
config=self.config)
File "/usr/local/lib/python3.5/dist-packages/tensorflow/python/estimator/estimator.py", line 1199, in _call_model_fn
raise ValueError('model_fn should return an EstimatorSpec.')
ValueError: model_fn should return an EstimatorSpec.

TBC. Tks.